dsh-knj-obsidian

Insights, gotchas, and design decisions that surface during AI agent conversations tend to evaporate the moment a session ends. dsh-knj-obsidian gives DSH agents a full lifecycle for a project-level knowledge base (.wiki/) stored in plain Markdown with YAML frontmatter: distilling conversations, documents, and web pages into structured knowledge pages that can be searched, maintained, and reused over time.

The write side offers incremental ingest with content-hash deduplication, quick capture of discussion conclusions, and lint health checks that surface orphan pages, broken wikilinks, and missing frontmatter. The retrieval side ships both a wiki_query tool and a wiki-query skill in a dual-channel setup, walking through four escalating layers from index fast-path to graph-neighbor expansion, with every synthesized answer carrying page citations and a confidence tag (extracted, inferred, or ambiguous). The graph side exports a zero-dependency interactive SVG force-directed visualization or structured JSON in a single call, rendering broken links as red dashed edges and orphan pages as grey nodes so structural issues are visible at a glance. The UI side adds a sidebar knowledge panel with a category-grouped tree, layered search reusing the same retrieval kernel, an embedded graph view, and a note workbench featuring rich Markdown rendering, in-place wikilink navigation with a history stack, source-code view, and full-file editing with path-traversal and cross-origin safeguards. Multi-vault management, batch Markdown import with idempotent re-import, and session distilling via a bundled wiki-distill skill are built in, and every visual element follows the host design tokens so light and dark themes adapt automatically.

Built for developers who work daily in DSH and want their agents to accumulate sharper context about the project over time, and for small teams that need to consolidate knowledge scattered across docs, chat logs, and internal wikis into a single repo-tracked, maintainable asset. The vault lives at the project root, travels with the code, and requires no external database or service.